#7 · ML / AI methods

Deep learning and LLMs prioritize antimicrobial peptides and extremophile bacteriocins

The dominant thread is computational prioritization of antimicrobial candidates directly from microbial genomes and chemical datasets. SEGMA, a structure-aware deep learning framework, mines 3,298 novel extremocins from 60,461 extremophile MAGs and then optimizes them. IIBacFinder, a genomics-guided pipeline using pHMMs and leader-peptide inference, mines roughly 280,000 human-gut genomes for class II bacteriocins. A separate study applies traditional machine learning and commercial large language models to antibacterial data for synthesized cyanine molecules, identifying positive charge among its structure-activity findings. What is new is the combination of deep learning and LLM strategies with specific compound classes. Other items, including the multimodal ACCESS annotation framework, the multi-round protein selection analysis, and the overlapping-gene design work, address adjacent problems rather than antibacterial candidate screens.